!!"
"#
$%&'(&&&'(&)*()
+%&%(&%
,-./0..&-/1..#233
4*-./..-/5%..#333
6(&7&7&((&&*&
%%&+&%&'(&
89*-.+7*(&*,:;
&'&%%&'(&%&<,:%
<&(%&*7*&&
%&'(&)+
.9*(%7&(&;+&%&'(&
9*(%&(&;+&%&'(&
==
>&&%)&
*&;&&7&&&%
(%7
=/>%&)&%**&
)%7&(&
/!%*+())'*)?
&&)&77&
'(%(*'&5%?
*'%@;(('7A""&'%)
%)%)*
A'/
%&%&)
'7&&%&&7
)%&7&
B /C%/
/!&/
/!DD/
-7/ %)*&&
**
EB /E7("'(("'%
;)((*'((
B A7
B=/3&7B(
7'.>&*B**(*&7
%&&7(*&7
%*=>&B*=(
&&&&7
F%2*7B/9D;>;%
G&D);
97B%%73&
)%'*&)&D
2%/
%2%9HF
I;;J";*"0&&%
A> #core pts/
border pts/()%%
FNoise pts
33K?*&((&
'%'(
F33K ?*%(&'
((
>LF29%)3FF&&
7)''7&*&&
=%0.,7;%.,%);%1.,(
(& /F%;%3
(&/!;
A&9(/>%*#+7*';
()*
A'/
F%2*7
Data Leakage
9%737)*%
9%733
*M'%
#(*%
%7'N%D7
&&%')NL7**%%*&%
$ED7%7(7#72%&%)
('7>(('7((7&5%&
%&'&7)%&
Detecting Data Leakage:
9*(*'7
&D*&7&)*%7)%7C>O
>*7**%7&'&'
Fixing Data Leakage:
C"7'7*%%7*%77
L7
>(('(7('%)
$;%%**(%7**%%
Making Model Robust to Outliers
mAP Mean Average Precision
E7EF7
!%')&D%7'*&
*%*%&&7&>
)%))'%7F7%JEE
3&*%%(&7/
E'/3&&()&*P
-'/3&*%&*'%*
9(/2)(*&&
'/'*"%*;P')%
9M'(%(%&%*
&*%G&*M&(%%&
%*&*%
>+%F5F5&)>+
F5D%&)E7>+F57
%((=!&=!
%)'()*&'((&&
'((!7!"J7
3&(&%7&&&&7"J7((&%7&
%((
Goal is to increase the margin.
Works fine for large datasets.
Also when features> data points.
Non-Linear Data also using Kernel Trick.
x- Wrong choice of kernel.
x- Extremely slow in test phase.
x- Complexity more.
22F*&;*2>
2>&%&*%7)'*
7&*%*
A'*)/Q)%;7R
A'*)/)%Q;7R
A'*)/7Q;)%R
=!%*&
E7(7&%7*'7(
)?)
>7)&)%
3
9*M/('*&)%)*(7%&
('*(79&%**7*D7
&%*)%
M(%'C('%7&
&((*%)5%(
M(%'%)%&())'**'7)
M(%'.0);.,0
Gradient Boosted Vs Decision Trees
K-Means ALgo
E/
%7%7(*%;)%&'5%
D7&J;&&()%'
(&7''J
39%(**5%&*5%'7
&())'**7"&%*5%'
&%*&&&%
9%&*5%&&
(%ST(
3&&*5%"
(*&(%
F7%J/
EE/%&*&7%J;&
%*7&)%%&
%D&7&(
3&*;%*75%"
(2%&&***&
%***%(%9)&%7&*
)&5%&7
>%7
$C'((7&(*&77
;'(&7&
GD&!&';&;>*%*&(
U(&FEL*%
**)%.;)%&('&.
*7%*(%3&T&'&FEL*%
E
>7(7&&%&7&"(7"
&(*%D7
7%%'**&7*&(%)&
7J3&%&(%J
=&7M/)'%7%;)&J
C"(7M/)'7((7;"J;
7&(*%)3&''
&&2'&D)%77&&
&'&((9*(7)7*%;
&'((*%7)
Multicolinearity :
!%'')()-;
%'())%&*&
**D&'
&7&The result is that the coefficient estimates are
unstable and difficult to interpret!%'(&
(*&';%&**&7;
D**%(*'&
=9*/
=**7*%(3&%%*=9
'%*&(7&&
5%**&**"(;=9*<'%&
&*(%**<.:)77&&'%%
"(*&%'V*&&&
(
A,'
M&,&7&'
E3&
A=3**
Normal Distribution Properties
Resampling
Types of Biases :
A#((*&((%
L7))*&((%5%'
)%
%&()7**%7(&%((
%7(%'D7&&
D)%*&D*(;
C"%7'%D+%&%)%;7
&&'(*7('-;&'(*)&&;
&&*;&%)7*&
G&'JN
C"(G7&>7
M&% F!(
>M&%KF!(
A& JN
*"%
Principal Component Analysis :
- Dimensionally Reduction while retaining the existing variance in dataset.
- transforming the variables to a new set of variables, which are known
as the principal components.
- Principal components are the eigenvectors of a covariance matrix, and
hence they are orthogonal.
- Eigen vector of a matrix A is a vector represented by a matrix X such
that when X is multiplied with matrix A, then the direction of the resultant
matrix remains same as vector X.
SVM Hard Vs Soft
-97E%
3&*7&*7*%;&&%%
&;'&%(%*&*
'".;W '⋅
2%(.,0'".;W .,.,⋅
2%.(.,0'".;W .⋅
Gini Index Calc
Entropy Calc
2%3
&7&
J&*%;&77&)%)%&7&%
E7*&*%&'%
$(&%
,&5%%*&
BC)%/G&%%*5%=%
B C)%/C=%
Cross Validation for Time Series data
!G7&/
9J.#;%'(**%
F'J.
RNN- Recurrent Neural Network
F 7D*(%&(J
E3!/
▹ 3&%7'*&DVD&
▹ 3&%(%&(%(VD&(%&
▹ 3&(%&%(
&7'(E3!?*77;(%7%(%7
7M/&&)*&7'*&
D%*%7)&&(%&(%
3&;&(%&&(%*
%D&7%(%
!' D/&(%&(%
7S'%(T
9(%M/*'&&(*&S'T&
73&D%(%*%Q.;R
$2%(%M
Autoencoders :
>%%DL%(!&7
7&&(()D((7;7&7%)5%
&(%>%%%&J*%(%
(9*'&7;&'%
*&(
9*'F%;->%***>N
>%*&
*%%('
9T&'9%%'
&&)*;7
9**'&%&&
&%7*&>
>%((*&'&%(%
9D%*('*&(('*
7&&
Image Coloring/')D&(%7(7
&&(%;()&&&%)
Feature variation:C"'&5%*%*77&
%(%)'7'%'%(
Dimensionality Reduction: 3&%7&%(%)%&
%9&((77&%("%
Denoising Image:3&(%)'&%&(%)%
&'%(>7%&%%
&7(%*&'
C/3&(*&D(&(%((
3&'&(%7((%
3&(7&*&77
/F(&((%&&*&;&&(*
)*&()
/&7)D&73&7
'%*&77%*&(
EF7/
1. Regression Coefficient :
2. Intercept (or β0):
3. Error Terms or Residuals (ϵ):
EF7(/
EMC5%/
%/Σ(%
**&*%
.
$M
,G7&L(
Building a Linear Model
Residual Analysis
F!C'%&!C%&(**&7
&&&&%&()
!C**)*%&D'(*&(
(**)*%D!>C
!>)%C!>C**)("(&(7
%5%%7%9'*(7%5%
%7%*(&7&&
&7
3&%*!>C;!C;F!C(&7&%'*7
-;&7&%*F5%)
3&F!C&7(&%*(
))%&F5%&&()
"(&&()
!C'))%&!>C;&5%7*&*
&7&(%
G&'EF7***N
3&*&EF7&%%%&
*()%
3&()77&&*&&%&(
MSE vs RMSE vs MAE vs R
2
vs Adj. R
2
G&'%77&&'7N
9*'%%7;&(%)7&
&J*'%*%7&&"%&%&'
)
2*&%(*EF&&*P%*
&'3&)&&)')G&,.(*
&((;&&,;"%%>
"%;&G&.;&X<.<;
((&J;&((&J
3&7*7*&YZ)YZ7&%(&%(&*
(PP@')%A%P'D&%.;&
%(('&+%*';(("'3&*;&*&
77&%(*'%%7&)'=
LOGITS:
(*&&7&E@'&&())'*)77
&7&<.3&&*)7%) .<.<.<.<
'2*)7....<<.
3&''%((7;)%&*)7&%)&((*&
*)7*GD*&''&%7&&%7&;
2*!<6
2*)7.6
&7*)7"'((&7*)7*
977;&()7;&&&%7*&;&;
77**%*
7K);&77%)';%()
&@'7&&G%&''7&7
%=;)%*%;@&7&;&.
E7))/
7K)
3'(*)%/
Bernoulli Distribution:>A%)%&'()%;
'%.*%;7
3&())'*%7)'/(
"
(
"
&"[.;
Uniform Distribution: G&'%*;&%43&
())*77&%5%'D'&&)*
%*)%LDA%)%;&%)*()%
*%*)%5%'D'
Binomial Distribution:
(
5
(%;5*;;*(*%
C&(
3&'()%&%*%
>%)*%
3&())'*%*%*3
$Normal Distribution:
3&;*&)%
3&%*&)%)&(')%&"\
3&%&%
C"'&**&%&**&&&&*&7&
,Poisson Distribution:
)%(()%&%(*
(&%'&%)*%*&
]&&&%;
&7&*;
0^]X
>&%)*&
4Exponential Distribution:
C"()%&*)&;7&*)
!;**>C"()%'%*%'
&"(**&&"(**&%
&(]0.&&&
How IQR used in Time Series?
9HF%3>_%_(&,&
9H%F79HF*&H%&!%%*&
3&H%&!"%
Data Preparation Operations in Time Series?
7**%%*
M75%**"*5%'(
!(%77&J*
F(77(%*5%'
*7&&)%
7G!&3
3&(%()%(%);&"(
)%&%(%)
3&%)*(%(&&J*&7
Trend : -7&&&7&)
Cyclical: -"&"&)*&**"(
Seasonal: "&"&)7%*%%)&;
''**"D(
Irregularity: L"((D&(
'/
!=%&%)
=&%)*
!%**)((
Augmented Dickey-Fuller (ADF) Test or Unit Root Test:
Detrending:
97&***&7&7
'&**%*&'&'(
)*
1. Differencing:%&%(%&**)
&7)&)&(%(
%))
3&&&***7*
((&D*&
9*&5%&&7&;&
***&'**)D;*
**73&()*%&(*
L7E!/%)#(
Transformation/3&%&**&&'3*;
5%F;E73*;'%E73*
Auto-Correlation9%&7*&')7
&77*&**&)
3&%*'*&7&*%**
)%&%')%
>%F7!/3&(;&(*%%(*)(
(*'%**7;&&)%
7&%&(%)D*&
Moving Average Methodology
(!7>7%7&*&(%! (3&*
7(;(7&%*&7(*
7&%*! ;(&&7&"(*%'
%%!7>7%7&*(%;&%
C"(!7>7*'*%3&7&*
7%'3&977&(;&
(&!>;&C!>*&7
Decision Rules in Hypothesis Testing
>'%;*7$;&>**
7*'3&*;
%%7(*$
;%&*&
**%7&P%GD5%/
Akaike’s Information Criterion (AIC)
3&%*%*7&(;;5*>F9!>3&>9"(/
&&&>9(&
Ljung-Box Test
3&*(*7%'&%9';&%)
&;&%
>7'&%&E+%7A" &&')((&
F%
3&%&'(&*&E+%7A"&&%ED7&
(%);&&')..,3&*;+&%&'(&
&%
>F9!>!(;;5/
(&%)*%7;
&%)****';
5&%)*77*&(5%
HOG – Histogram of Oriented Gradients
G&'K&7N
-!D!/
.#E;(;&;5%*
/`;;;a
2)/`%;*%a
())'/`bU/.;bU/.;a
3))'/`bU/`bU/..;UU/.;a;bU/`bU/.;aa
C)/`bU/`b%/.;*/.<a;bU/`b%U/.;*/.6a?a
L7(&&((%')%
%
L73()(&)&"7
E&3"H%'M
/((&))
75%7"
>EF&7
-J&!99>J%7
D7
>*M
